Profiler-Guided Detection and Refactoring of Runtime Hotspots in Java Applications

Runtime hotspots in Java applications are usually chased down by intuition rather than by measurement. This study profiles three applications chosen to represent three structurally different hotspot categories: a Sudoku backtracking solver that is algorithmic and CPU-bound, a headless Snake simulation that is allocation-bound, and an invoice generator dominated by string building and file input/output. JDK Flight Recorder is the only profiling instrument used, and each hotspot receives a refactoring matched to its category. Execution time is measured independently across ten iterations per configuration and improves significantly in every application, by 38.7 percent for the invoice generator, 52.1 percent for the solver and 67.4 percent for the simulation. Peak heap, allocation rate and garbage-collection pause time are drawn from a single aggregated recording per configuration, so they are presented as descriptive point estimates and carry no significance test, and CPU time is reported as a proxy equal to execution time because no independent instrumentation was implemented. The invoice generator improves on all five metrics and rests on the largest volume of underlying events. The other two applications each yield one result opposing the direction their refactoring targeted, which the present recording design cannot separate from sampling artefact. Every refactoring was validated against byte-identical output before measurement began. The work shows that a profiler-first workflow reliably identifies and removes hotspots, while also demonstrating how measurement design determines which performance claims a study can legitimately support.